N COMMEN

Detecting DNA not from a organism such as a dinosaur, but rather from an extant organ­ ism, most likely a human. Determining the authenticity of an an­ The fact that DNA sequence can be ob­ DNA sequence (2) with human (Fig. O. cient DNA sequence often can be diffcult, tained from organisms has opened Although statistical support for most and criteria for this have been discussed else­ new windows of opportunity for research in nodes in the tree is low as a result of the where (1, 9). Two criteria that are impor­ organismal and molecular (1). short length of this region (133 base tant, and that were not fulfilled in the study Among these is the possibility of obtaining pairs), bootstrap support for this cluster by Woodward et al., are phylogenetic con­ genetic information from major groups of (91 %) is relatively high. Furthermore, a text and independent replication. Although organisms now extinct. Recently, S. R. consensus sequence of the nine bone se­ phylogenetic support has been presented for Woodward et al. sequenced DNA from a quences which maximizes similarity to hu­ other findings of DNA surviving for millions portion of the mitochondrial cytochrome b man (118/133 = 88% similarity) clusters of (10), real advance in this field will gene from Cretaceous bone fragments ap­ with the human sequence at a statistically come only when it is demonstrated that parently from a dinosaur that lived 80 mil­ significant bootstrap P value of 100%. those studies can be replicated in indepen­ lion years ago (2). However, the likely Consensus sequences with similarity max­ dent laboratories. source of those DNA sequences appears to imized to each of the other taxa yield S. Blair Hedges be human contamination. considerably lower (0 to 46%) probabili­ Department of Biology and In addition to experimental controls, a ties for clustering with the to which Institute of Molecular Evolutionary Genetics, major line of evidence normally used to sup­ similarity was maximized (8). 208 Mueller Laboratory, port a finding concerning ancient DNA is Despite meticulous care, contamina­ Pennsylvania State University, the phylogenetic relationship of the putative tion of polymerase chain reaction (PCR) University Park, PA 16802, USA ancient sequence to those from the closest experiments with foreign DNA, often of Mary H. Schweitzer living relatives of the fossil organism (1). In human origin, is an ever-present aspect of Paleontology Department, the case of a possible dinosaur sequence, ancient DNA research because of the sen­ Museum of the Rockies, there is strong evidence from morphology sitivity of the methodology and rarity of Bozeman, MT 59717, USA, that represent the closest living organ­ the target molecules (1). The suggestion and Department of Biology, isms to , and morphological and by Woodward et al. (2) that variation Montana State University, molecular evidence indicate that crocodil­ among the nine sequences (seven from the Bozeman, MT 59717 ians are the closest living relatives of birds same bone fragment) is a result of damaged (3-4). Also, the fossil record indicates that, template may be correct. However, our REFERENCES AND NOTES after splitting with mammals, at least 100 results suggest that the DNA template was 1. S. Paabo, R. G. Higuchi, A. C. Wilson, J. Bioi. Chem. million years of evolution occutted on the 264, 9709 (1989); S. Paabo, Sci. Am. 269, 86 (No­ lineage leading to dinosaurs and birds before vember, 1993). 2. S. R. Woodward, N. J. Weyand, M. Bunnell, Science the latter groups diverged (3). Therefore, a ____~9~1r======c;,Human I Cretaceous bone 265,1229 (1994). putative dinosaur sequence would be expect­ 3. M. J. Benton, J. Mol. Evol. 30,409 (1990). ed to cluster with birds and crocodilians in a 4. S. B. Hedges, Proc. Nat!. Acad. Sci. U.S.A. 91,2621 phylogenetic analysis of amniotes. '------Rabbit (1994). Rhinoceros 5. BLAST version 1.4.7MP [So F. Altschul, W. Gish, W. Woodward et al. (2) do not present an ,----- Dugong Miller, E. W. Myers, D. J. Lipman, J. Mol. Bioi. 215, evolutionary tree, but discuss their sequenc­ 48 Rodent 403, (1990)]; some sequences were of multiple indi­ es in terms of percent sequence difference, '----- Whale viduals of the same species. noting that these cytochrome b sequences '------Cow 6. Sites 15646, 15687, 15703, and 15706; S. Ander­ L------::-:i--- Sicklebill son et al., Nature 290,457 (1981). differed from all others in the databases. We ,------Shoebill 7. The DNA sequences were analyzed with MEGA ver­ also performed a BLAST search using the Domestic sion 1.01 [So Kumar, K. Tamura, M. Nei, MEGA: Mo­ majority rule consensus sequence [figure 6 in '-----Cuckoo lecular Evolutionary Genetics Analysis, (Pennsylvania State University, University Park, PA, 1993)] for dis­ (2)] and obtained matches to 130 cyto­ tance analyses and PAUP [D. L. Swofford, Phyloge­ chrome b sequences of vertebrates (5). As 0.1 netic Analysis Using Parsimony, Version 3.1 (Univer­ reported by Woodward et al. (2), the con­ Fig. 1. Phylogenetic tree of partial cytochrome b sity of Illinois, Champaign, IL (1993)] for parsimony analyses. Average pairwise JUkes-Cantor (T. H. Jukes DNA sequences in representatives of extant sensus sequence differs by about 30% (26% and C. R. Cantor, in Mammalian Protein Metabolism, to 52%) from those vertebrate sequences in tetrapod groups and putative dinosaur DNA se­ H. N. Munroe, Ed. (Academic Press, New York, 1969, the databases. However, 87 of the most sim­ quence (majority rule consensus) derived from Cre­ pp. 21-132) corrected distances were large (0.3 to taceous bone fragments (2). Numbers on nodes 0.5), and therefore a transversion distance (M. Kimura, ilar sequences (closest matches) are mam­ are bootstrap confidence probabilities. Inclusion of J. Mol. Evol. 16, 111, 1980) was used with neighbor­ mals, including all nine eutherian orders all nine putative dinosaur sequences (2) resulted in joining [N. Saitou and M. Nei, Mol. Bioi. Evol. 4, 406 represented, whereas birds, amphibians, and (1987)]; and transversion only, or transversions an identical tree in which those sequences clus­ weighted 10 times transitions, were used with parsi­ comprise nearly all of the remaining tered together with human. A frog was included to mony. The marsupial sequence was exciuded from sequences and have the lowest similarity to root the tree. Tree shown is neighbor-joining the phylogenetiC analyses because of anomalous re­ the consensus sequence. Among the mam­ with transversion distance; parsimony analyses sults. Statistical significance (>95%) was assessed (transversions only and weighted transversions) with the bootstrap method [J. Felsenstein, Evolution mal sequences, the closest matches are to 39, 783 (1985)], with 2000 replications. whales (99/133 = 74% similarity). Howev­ also clustered the putative dinosaur DNA sequence 8. Bootstrap probabilities for clustering with the taxon­ er, among the nucleotide sites showing sim­ with the human sequence; 133 sites total, 88 specific maximized consensus in neighbor-joining variable, and 66 parsimony. GenBank accession analyses (transversion distance) are as follows: hu­ ilarity to the human sequence (93/133 = numbers: human (V00662), bat (L28943), rhinoc­ man (100%), bat (17%), whale (8%), rabbit (46%), 69% ), four are rare variants in the other 129 eros (X56283), dugong (U07564), cow (J01394), dog (6%), rhinoceros (31 %), cow (22%), dugong vertebrate sequences (6). dog (L29416), rabbit (U07566), whale (X75581), (9%L rodent (26%), sicklebill (1 %), domestic fowl A phylogenetic analysis (7) with all (21 %), cuckoo (0%), shoebill (1 %), and frog (6%). rodent (L 11902), sicklebill (X74253), domestic fowl 9. T. Lindhahl, Nature 365, 700 (1993); O. Handt et al., tetrapod sequences obtained from the (X52392), cuckoo (U09262), shoebill (U08937), Experentia 50, 524 (1994). BLAST search joins the putative dinosaur and frog (U02890). 10. E. M. Golenberg et al., Nature 344, 656 (1990); P.

SCIENCE • VOL. 268 • 26 MAY 1995 1191 S. Soltis, D. E. Soltis, C. J. Smiley, Proc. Nat/. taxa with high alignment scores. Reducing ment two with the human and next with the Acad. Sci. U.S.A. 89,449 (1992); R. J. Cano, H. N. Poinar, D. W. Roubik, G. O. Poinar, Med. Sci. Res. the high PCR failure rate (1) in this way other unknown fragment. This resulted 20,619 (1992); R. DeSalle, J. Gatesy, W. Wheeler, should greatly increase the amount of se­ when the characters for each codon were D. Grimaldi, Science 257,1860 (1992); R. J. Cano, quence available for phylogenetic analysis. numbered and third positions were omitted H. N. Poinar, N. J. Pieniazek, A. Acra, G. O. Poinar, Nature 363, 536 (1993); H. N. Poinar, R. J. Cano, Steven Henikoff and when we looked at the more conserved G. O. Poinar, ibid., p. 677. Howard Hughes Medical Institute, transversions. When amino acids were trans­ 11. We thank C. A. Hass, S. Kumar, and S. Piiiibo for Fred Hutchinson Cancer Research Center, lated from the original nucleotide sequences helpful comments. Seattle, WA 98104, USA and parsimony analysis was conducted, the 6 December 1994; accepted 7 April 1995 unknown fragments were closest, then the REFERENCES chicken (supported by two characters). This The comparisons reported by Woodward et pattern also resulted when all characters 1. S. R. Woodward, N. J. Weyand, M. Bunnell, Science al. (1) were limited to identity percentages, 266,1229 (1994). were examined. whereas more informative comparisons 2. S. F. Altschul, J. Mol. Bioi. 219,555 (1991). Our most conservative and informative should be possible by scoring each aligned 3. S. Henikoff and J. G. Henikoff, Proc. Natl. Acad. Sci. analyses point to mammals as the closest amino acid pair with the use of a log-odds USA 89,10915 (1992). relatives to the available "Cretaceous" se­ 4. D. T. Jones, W. R. Taylor, J. M. Thornton, CABIOS 8, substitution matrix based on homologous 275 (1992). quence, an unlikely relation if these are protein alignments (2). I compiled a data­ 5. S. Henikoff and J. G. Henikoff, J. Mol. Bioi. 243, 574 truly dinosaur remains. This contradicts base of all 223 cytochrome b segments from (1994). with numerous morphological characters 6. U. Amason and A. Gullberg, Nature 367,726(1994). different species in the combined protein 7. S. Henikoff, unpublished data. that support birds as the closest living rela­ databanks (through 11/94). Each segment tive to the dinosaurs (3). One might ask, was scored for similarity to a consensus rep­ 9 December 1994; accepted 7 April 1995 how did mammalian DNA get into these resenting the seven long bone sequences, samples? At the time that these coal beds with the use of the most frequent predicted Assuming that each of the published se­ were formed, all of the known mammals amino acid at each position. A range of quences (1) are representative of the study were smaller than the bone fragments de­ BLOSUM (3) and PAM (4) substitution by Woodward et aI., we chose two for scribed (4). Possibly, either ancient DNA of matrices was used for scoring. In addition, extensive analyses to assess the history of a smaller mammal was preserved along with each segment was scored using position-spe­ these molecules (3-37 from bone fragment these deposits and thus contaminated these cific scoring matrices (5) constructed from one and 5-37 from bone fragment two). bone fragments, or a more recent DNA the seven long bone sequences and from the When alignments were determined by sample contaminated these tissue samples. two rib bone sequences. comparison against all of the sequences in Fossil mammals are known from this geo­ All tested scoring systems provided similar a current issue of Entrez (NCBI, release logical formation, potentially supporting results (data, not shown). Among the well­ 6.0 of GenBank) with the use of the the former hypothesis. We prefer the latter represented' taxa, the highest mean scores MacVector program (version 4.1.4, East­ hypothesis because of the great similarity of were found for cetaceans and ungulates. In man Kodak, Rochester, New York), the these sequences to living mammalian genes. both cases the mean scores are significantly best 30 alignments against fragment one Marc W. Allard higher than the mean scores for birds. It is were all mammalian cytochrome b se­ Deshea Young notable that all 15 alignments with cetacean quences, with the first nine chosen from Yentram Huyen segments outscored all n alignments with the order Artiodactyla (cattle, deer, ante­ Department of Biological Sdences, segments, even though both groups are lopes, and their relatives). A similar result George Washington University, diversely represented (6). Overall, scores for was obtained for alignments against frag­ Washington, DC 20052, USA vertebrates were much higher than for arthro­ ment two, with the best four alignments pods, which in tum were much higher than each to human cytochrome b genes. Other REFERENCES for non- (plants, fungi, and bacteria), vertebrates are not equally divergent from 1. S. R. Woodward, N. J. Weyland, M. Bunnell, Science indicating that this method applied to bone these purported dinosaur sequences. To 266,1229 (1994). sequences provides rankings consistent with the contrary, these unknown sequences 2. R. DeSalle, J. Gatesy, W. Wheeler, D. Grimaldi, ibid. known phylogenetic relationships. Moreover, have closest similarity to the mule deer 257, 1933 (1992). similar results were found for rib sequences (Odocoileus hemionus, accession number 3. J. A. Gauthier, Mem. Calif. Acad. Sci. 8, 1 (1986). 4. J. A. Lillegraven and M. C. McKenna, Amer. Mus. analyzed independently of long bone se­ X56291) and. to human cytochrome b Novitates 2840 (1986). quences, despite several nucleotide sequence genes (Homo sapiens, accession number differences (1). V00662), respectively. 19 December 1994; accepted 7 April 1995 I conclude that the bone sequences more The best strategy for determining related­ closely resemble homo logs in mammals than ness of an unknown sequence is not through Our preliminary phylogenetic analysis of in birds, which are thought to be the closest a similarity search, but rather by a phyloge­ the putative dinosaur sequences in the report living relatives to dinosaurs. Furthermore, netic analysis using parsimony (2). While we by Woodward et al. (1) showed them to be the significantly higher scores for some agree with Woodward et al. (1) that their weakly related to the human cytochrome b mammals (cetaceans and ungulates) than for small fragment of cytochrome b sequence is gene, albeit quite distantly (earlier comment others (7) further suggest either a mammali­ inappropriate for use in a phylogenetic anal­ by Hedges et aI., data not shown). As nuclear an origin or convergence of this region of ysis, it is the only available evidence, and insertions of mitochondrial DNA are known cytochrome b. The analysis also contradicts parsimony is still the best strategy for deter­ to occur (2), and as 12S ribosomal DNA criticisms that the bone sequences resulted mining the closest relative and for identify­ sequences amplified from ancient monkey from microbial contamination or were seri­ ing these new sequences. We aligned the bones have been attributed to insertions of ously affected by PCR-generated errors. unknown cytochrome b sequences to several mitochondrial DNA into the human nuclear Therefore, further PCR-based analysis of the mammals (human, cow, rat, and mouse), to genome (3), the putative dinosaur cyto­ Utah bones is warranted. For such studies, chicken, and to clawed frog (Xenopus, our chrome b sequences might represent ancient most efficient synthesis should be possible outgroup). The most parsimonious solution integrations of mitochondrial DNA into the with primers modeled on the mammalian was one that grouped Cretaceous bone frag- human nuclear genome.

1192 SCIENCE • VOL. 268 • 26 MAY 1995